External information environment required for cross disciplinary thinking in large language models based on...

Area of Science:

• Artificial Intelligence
• Cognitive Science
• Information Science

Background:

• External information stimuli are vital for human knowledge reintegration and cross-disciplinary thinking.
• Understanding how information environments influence AI, specifically Large Language Models (LLMs), is crucial for advancing AI capabilities.

Purpose of the Study:

• To investigate the effect of the external information environment on LLMs' cross-disciplinary thinking.
• To analyze how different information contexts influence LLMs' ability to generate interdisciplinary lay summaries, assessing convergent and divergent thinking.

Main Methods:

• Utilized general abstracts and lay summaries to create deep and broad knowledge contexts for LLMs.
• Compared intersection interpretations and cross-inspirations generated by ChatGPT-4.0 and Claude3 under varied information input combinations.
• Assessed LLMs' convergent and divergent thinking through interdisciplinary lay summaries.

Main Results:

• Simultaneous access to deep and broad knowledge contexts enhanced LLMs' convergent thinking.
• Intersection interpretations within the information environment fostered divergent thinking in LLMs.
• A modest correlation was observed between the external information environment and LLMs' divergent thinking.

Conclusions:

• The external information environment plays a significant role in shaping LLMs' cross-disciplinary thinking capabilities.
• Tailoring information input can optimize LLMs for specific cognitive tasks like convergent and divergent thinking.
• This research contributes to advancing the application of LLMs in interdisciplinary research and knowledge synthesis.